Arguably one of the most challenging aspects of studying PD is the evolving nature of its clinical definition (Figure 1) which has complicated epidemiological efforts to estimate prevalence and study trends over time. Pregnancy-related depression was not distinguished by specifiers or a unique diagnosis in the Diagnostic and Statistical Manual (DSM) until 1994 when it appeared as a postpartum specifier to MD in the DSM-IV (American Psychiatric Association, 1994). The most recent edition (DSM-5) expands the definition of PD to include any major depressive episode that onsets either during pregnancy or within the first 4 weeks postpartum (American Psychiatric Association, 1994). The International Classification of Diseases code (ICD) recently underwent a similar expansion from ICD-10 and ICD-11, to include the prenatal period and up to 6 weeks postpartum (World Health Organization (WHO) 1993; World Health Organization (WHO) 2019). While this shift in clinical definition reflects improvements in prenatal care, the need to harmonize data across multiple diagnostic definitions complicates meta-analyses and increases difficulties in drawing cross-study comparisons (Rantalainen et al., 2020; Bauer et al., 2019).

A challenge related to but distinct from defining PD is measuring it. A variety of measurement strategies exist, and choice of measurement tool(s) directly impacts the cost of research and the interpretability of the results. Broadly speaking, psychiatric disorders can be assessed by diagnostic thresholds or by symptom-based measures. Symptom-based measures typically correlate with clinical diagnosis status and may index disorder severity, but they are not tantamount to clinical diagnoses, which typically yield greater heritability estimates and smaller prevalence estimates (Kendler et al., 2019; Foley et al., 2001). The most commonly used measurement tool in PD research is the Edinburgh Postnatal Depression Scale (EPDS), a validated self-report questionnaire measuring PD symptoms (Cox et al., 1987). The EPDS was designed to prioritize mood symptoms over somatic symptoms (e.g., feelings of sadness versus changes in weight) to prevent physiologically normal changes in weight, appetite, and energy that are typical during the peripartum from inflating the depressive symptom count. The questionnaire is short, simple to administer, and sufficiently accurate as a clinical screening tool (sensitivity and specificity estimated as 85% and 84%, respectively, using a cut-off of 10) (Levis et al., 2020). The EPDS is validated, but not in every language or culture. As psychometric studies suggest that the accuracy of EPDS scores varies across countries, languages, and cultures, it may not be an ideal instrument for research outside of majority English-speaking countries (Halbreich and Karkun, 2006).

Minimal phenotyping (sometimes called broad or shallow phenotyping) refers to using short, often non-validated questionnaires to collect self-reported details about demographics, medical history, disorder status, and more. An example of minimal PD phenotyping would be the use of a single question asking, “Have you ever been diagnosed with perinatal depression? (Yes/No)” to determine PD case status. This phenotyping strategy would not collect details about the episode (e.g., timing of onset, duration, and severity) or a comprehensive account of the participant’s history of MD and PD. In contrast, deep phenotyping includes assessment tools like structured clinical interviews, which are the most accurate psychometric instruments for evaluating psychiatric disorders. Clinical interviews generate rich, reliable data but are time-intensive and costly to perform. Thus, they may be prohibitively expensive or infeasible to collect for large-scale studies. In general, the more measures collected, the greater the financial cost and accuracy of the data. A compromise between minimal and deep phenotyping is to use one or more validated self-report questionnaires and/or medical record abstraction.

Recent research exploring the impact of minimal phenotyping on efforts to characterize the genetic architecture of MD has found that using broadly-defined health outcomes instead of rigorously-defined phenotypes can produce misleading and non-specific results (Cai et al., 2020). Broad MD phenotypes derived from minimal phenotyping have significantly different genetic signatures from clinically-defined MD to the extent that some minimal MD phenotypes have greater genetic correlations with other psychiatric disorders than clinically-defined MD does (Cai et al., 2020). In other words, minimal MD phenotypes appear to index general psychopathology rather than MD-specific psychopathology (Kendler et al., 2019; Cai et al., 2020). These consequences primarily stem from the fact that the minimally phenotyped samples do not actually represent clinical MD (Kendler et al., 2019; Cai et al., 2020). Minimally defined MD cases (including PD) represent an aggregation of several separate phenotypes, including patients meeting clinical criteria for MD, participants with subclinical levels of depression, misdiagnosed participants who have related but distinct condition(s) (e.g., bipolar disorder), individuals scoring high for neuroticism, and patients prescribed antidepressants to treat off-label conditions (Kendler et al., 2019). This phenotypic heterogeneity obscures association signals and is a plausible explanation for why MD defined by shallow phenotyping is less heritable (SNP-based heritability of 11%–13% compared to 26%) and more genetically related to other psychiatric disorders than stricter clinical definitions of MD (Cai et al., 2020). In addition to contributing to challenges in identifying trait-specific molecular pathways, low resolution data also decreases the odds that responses and medical records will be comparable since the ancillary details (e.g., timing of episode onset in the case of PD) are not available to facilitate harmonizing data from multiple sources.

An analogous depth of phenotyping study has not been performed for PD because sufficient data does not exist. PD is underrepresented even in large, EHR-based population cohorts like All of Us (All of Us Research Program Investigators et al., 2019). As of 6 June 2022, the All of Us study included 372,380 participants. Of the 227,740 participants with EHR data, 1.4% had a diagnosis of schizophrenia (population prevalence of 1%) compared to 0.2% with postpartum depression (prenatal and peripartum depression are not listed). This underrepresentation echoes and is partially driven by PD going unrecognized in clinical settings. Even without the data necessary to perform a replication study, it is reasonable to assume that the findings from the MD depth of phenotyping study apply to PD because the phenomenon of genetic associations changing with respect to disorder definition has been observed for other disorders, particularly those at risk for frequent misdiagnosis. A 2022 review on the progress of Alzheimer’s GWAS research noted that heritability estimates had fallen from 17% to 3% when the cohorts transitioned from using clinically assessed Alzheimer’s disease cases to “proxy dementia cases” despite an increase in sample size (Escott-Price and Hardy, 2022). The high degree of clinical, and likely etiological, heterogeneity in complex disorders like MD and Alzheimer’s disease, may further exacerbate the consequences of minimal phenotyping in genetic research.

Comprehensive reporting, particularly on the timing of episodes and types of measures, can improve data harmonization, both by creating more opportunities for meta-analysis and assisting in interpreting results across studies. It remains essential to use accurate language to describe the periods surrounding pregnancy and childbirth (e.g., peripartum, prenatal, postpartum), so that depressive episodes occurring during pregnancy are clearly distinguished from depression outside of the peripartum period. Clarifying pregnancy-associated terms is especially pertinent as PD definitions have shifted over time to include episode onset during pregnancy (i.e., from DSM-IV to DSM-5 and ICD-10 to ICD-11; Figure 1) (World Health Organization, 1993; American Psychiatric Association, 1994; World Health Organization, 2019). Effort should also be made to avoid conflating PD symptoms with diagnoses. The majority of whole-genome studies of PD rely on symptom-based screening instruments, such as the EPDS, to measure PD (Byrne et al., 2014; Rantalainen et al., 2020; Pouget et al., 2021; Kiewa et al., 2022). These tools are used either as a means to determine case status by setting a symptom threshold or to test relationships between genetic factors and symptom load as a continuous measure. Given that distinct genetic architectures have been identified for clinical diagnoses of MD and depressive symptom measures, (Kendler et al., 2019) equating a clinical diagnosis of PD with symptom-based measures may introduce additional noise and reduce power to detect PD-specific genetic effects. Researchers should clearly and consistently distinguish between symptom-based and diagnosis-based phenotypes, and may consider evaluating results from studies examining PD symptoms separately from those using clinical diagnoses. In some cases, studies may collect more details than they share (e.g., collecting symptom-level data but sharing only case-control status), resulting in low resolution data and decreasing the odds that the data will be comparable. Openly sharing detailed accounts of all available measures with associated summary statistics can help address these challenges.

Considering that a well-powered GWAS of PD has yet to be published, increasing sample collection is a primary concern. However, larger sample sizes at the cost of phenotypic precision may not be the only or best strategy to detect PD-specific genetic effects (Escott-Price and Hardy, 2022). The China, Oxford and VCU Experimental Research on Genetic Epidemiology (CONVERGE) consortium identified the first robust genome-wide associations with MD despite having fewer samples than other large-scale efforts by focusing on deeper phenotyping and reducing the clinical heterogeneity of the sample (Major Depressive Disorder Working Group of the Psychiatric GWAS Consortium et al., 2013; CONVERGE Consortium, 2015; Peterson et al., 2018). Similarly, PRS generated using strictly-defined MD had the same predictive ability as broadly-defined MD but in smaller sample sizes (Cai et al., 2020).

Several attributes of PD require the need for accurate and comprehensive data collection. As PD is a subtype of MD, more detailed data collection may be necessary to understand the potentially subtle differences in variation between PD and MD. Collected cases need to be clearly distinct from both MD occurring outside of the peripartum and from women experiencing the “baby blues” to minimize the possibility of misclassification. Considering that the PD diagnostic criteria has changed over time (Figure 1), detailed longitudinal collection will allow for continued data harmonization and improve data reuse even if definitions continue to evolve. Details beyond PD case status, such as the timing of any MD episodes, pregnancies, and births, will need to be recorded to verify whether individuals continue to meet criteria. If following current diagnostic codes, the window of time postpartum to meet criteria for PD is relatively small (Figure 1) and may also benefit from prospective data collection. Finally, given the interest in identifying PD-specific processes and understanding how they may differ from MD etiological mechanisms, (Kiewa et al., 2022; Bauer et al., 2019; Byrne et al., 2014; Pouget et al., 2021) comprehensive information about participants’ history of depression and other comorbid psychiatric disorders should be collected.

A disproportionate majority (>78%) of large-scale GWAS of major psychiatric disorders are conducted in European ancestry cohorts (Peterson et al., 2019). Indeed, the largest GWAS of PD only included European ancestry individuals (Kiewa et al., 2022). The underrepresentation of diverse ancestral populations in research seriously limits the understanding of disease etiology. Historically, traditional GWAS approaches have required ancestral homogeneity among study populations, which has excluded ancestrally admixed individuals and underrepresented ancestral groups in order to reduce spurious findings from effects of population stratification. However, emerging trans-ancestry methods are increasingly available and should be applied to be as inclusive as possible (see Table 1) (Peterson et al., 2019). Given that large-scale PD data collection efforts are required, the field and funding agencies have an opportunity to make deliberate data collection choices. Inclusive recruitment strategies will require significant time devoted to trust building and community engagement such as in community-based participatory research (Cargo and Mercer, 2008). In addition to community engagement, research teams should include diverse perspectives and lived experiences including expertise in cross-cultural psychiatry as well as members from historically underrepresented groups in research (Hofstra et al., 2020).

Large-scale psychiatric genetic studies have largely ignored important environmental risk factors in their models despite known effects. This practice limits the understanding of disease and leads to incomplete etiological models. Including environmental risk factors can aid in detecting genetic associations by accounting for heterogeneity in risk. For example, in MD, additional genetic risk variants were uncovered by stratifying participants in the GWAS by severe adversity exposure (Peterson et al., 2018). Similarly, data collection efforts should include metrics of social determinants of health (e.g., access to prenatal care, social support) in order to identify interactions with genetic risk and points for intervention for improved health outcomes. Race and ethnicity are socially-derived categorical constructs that can provide information about social factors that affect risk for disease, including having a lived experience of social injustice. However, they are often inappropriately used as a proxy for genetic ancestry which is a biologically-derived continuous construct. Researchers should clearly distinguish race from ancestry in order to tease apart specific biological, psychological, and social determinants of health. Broadening ancestral diversity of studied populations coupled with investigations of social determinants of health will improve the effectiveness of genomic medicine by expanding the scope of known genomic and environmental risk and resilience factors towards advancing the understanding of PD disease etiology.